MMSUPER8

Author: Michael Minn (see michaelminn.com for current address)

July 25, 2004

Instructions for using the mmsuper8 utility to process scanned images of super8 film into digital video files.

1. Introduction

The following page provides instructions and tips for creating digital video files from super 8 film scanned with a flatbed scanner. While the quality is poorer than that achieved with a professional telecine machine, it is acceptable for casual viewing and archiving of super 8 home movies and commercial prints.

There are numerous advantages and disadvantages to this approach.

Time. The biggest issue is time. Scanning super 8 film with a flatbed scanner is extremely time consuming and tedious. After finding a comfortable and efficient physical setup (a nontrivial task), it should be possible to scan around 50 feet of film (3 minutes) per hour.
Cost. Professional quality transfer of film usually runs around $0.25 per foot. While this is not prohibitively expensive, it is also not insignificant. Flatbed scanners with transparency adapters are widely available and, compared to a $20K telecine, relatively inexpensive.
QualityTechniques and equipment used by transfer houses vary widely and you may end up spending alot of money for a poor quality transfer. By being actively involved in the process, you can adjust the image quality to suit your taste. While the quality of flatbed scanned film does not match the highest quality professional transfers, it may be acceptable for your needs.
Archive. Storage media and video compression codecs come and go with great frequency and there is no assurance that hardware or software will be available to recover your content 10 or 20 years down the road. Preserving content in a simple uncompressed format like PNM files requires more space than compressed formats, but stands less chance for obsolescence, provided that the data is migrated to new physical media along with your other important data on a regular basis.
Privacy. In submitting your film to a transfer house, you have no assurance that your intimate film moments will not end up on the internet. That naked pool picture of you when you were three can cause problems if the film technician feels like making it an issue.
Independence. Frees you from dependency on expensive technology to access your precious movie memories. It also allows you to transfer and edit your content without need for expensive DV hardware that you might not otherwise need.

There are a few other pages on the web provided by other people who have tried flatbed scanning of super8 film:

Jim Carroll's Page

8mm2avi

2. Version History

v2004-07-25: Debut beta release to the public.

3. Scanning

Sample scan from a commercial print

The challenge with scanning is to find a physical setup that will hold the film steady in the scanning area and make it possible to quickly advance the film between scans. As the interface between the computer and the physical world with inadequate equipment, scanning is the most problematic part of the process.

Scanner: The scanner itself should be at minimum capable of 1200 dpi OPTICAL scan resolution (not interpolation from a lower resolution) with 2400 or 3600 even better (but slower). I use an Epson Perfection 2400, a USB scanner supported under Linux and providing 2400 DPI resolution.

Light Source: You will need some sort of diffuse light source to backlight the film. Scanners that have transparency adapters built into their covers usually place them in the same orientation as the scanner glass. Although the results MAY be more accurate, scanning a high resolution lengthwise (using only a small portion of the scanning platen) is MUCH slower than scanning sideways (using the full length of the scanning platen). Therefore, some way of back illuminating the film must be found. I use an old freestanding HP ScanJet HPA transparency adapter that I purchased on ebay. To hold the film down on the glass, I remove the diffuser and light unit from the case to press the film against the glass. Finding something that will provide a consistent light source is difficult.

Film Guide: While the mmsuper8 software detects sprocket holes and crops individual frames, some kind of guide is needed to place the film in a consistent place on the glass and facilitate pulling an appropriate number of frames through between scans without and undue number of dropped or duplicated frames. Using foam mounting tape, I created guides on each side of the scanner. As noted above, scans are performed sideways, which is much faster than scanning lengthwise.

Scanimage Script: SANE (the Linux Scanner support package) provides the command line scanimage utility for controlling scanners. This permits using a script to control scanning and reduce the number of mouse actions and keystrokes that must be performed for each scan. This will require some tweaking for your individual situation to reduce the scanned area to the minimum area needed to capture as many frames as possible. The example given below also uses the pnmflip command since the film is scanned sideways, but mmsuper8 requires strips of frames that are oriented top to bottom.

	index=0
	while true; do
		filename=`printf "%03d.pnm" $((index))`
		echo "press  to scan" $filename
		read
		scanimage --resolution 1200 -l 15 -t 11 -y 11 -x 185 --mode Color > temp.pnm
		pnmflip -rotate270 temp.pnm > $filename
		rm temp.pnm 
		index=$((index+1))
	done
	exit

Although it is usually impossible to get the scanner to run without the scanner lamp on, this does not adversely affect the quality of most well-exposed film, provided that there is a consistent light behind the transparency. Dirt and scratches on underexposed film will reflect light, causing the defects to be light colored in the scan.

Scanners with built-in transparency adapters detect the presence of the adapter, so you can't use the transparency setting if the lid has been removed. Also they usually limit the scanning dimensions to the area convered by the transparency adapter, and as stated above, lengthwise scanning of super 8 film at high resolution is prohibitively slow.

To detect edges successfully, mmsuper8 requires that the scan strips be oriented top to bottom and that the left and right areas in the scan be white. The image should also be cropped so that the frames are usable from top and bottom.

With my physical setup, each scan captures around seven inches (40 frames - around two seconds) and takes about a minute. At this rate, I can scan a 50-foot roll in around an hour.

4. mmsuper8

mmsuper8 takes as input scans of super 8 film in pnm files. Files may contain single frames or strips of frames that are oriented top to bottom and have ONLY white area to the left and right of the film scan (for edge and sprocket hole detection).

mmsuper8 takes as input one or more pnm files. PNM is an extremely simple, uncompressed format that is the output of the scanimage program. While this format consumes a significant amount of space (around 200K/frame at 2400DPI), it is extremely easy to read, does not use lossy compression, and should be readable in the future when proprietary format algorithms have been long forgotten.

mmsuper8 can output individual PNM frame files, an uncompressed RGB AVI file or an uncompressed .mov (Quicktime) file. Also note that mmsuper8 can be used to combine individual PNM frame files into an .avi or .mov file later, but there is currently no utility to extract individual frames from a .mov.

You should try a few sample scans before putting in the effort to scan a whole roll of film.

	mmsuper8 [options] scan_file.pnm ...
	
	
	-o [.avi file] | [.mov file]
		Sends output to uncompressed .avi or .mov (Quicktime) file.
		By default, output goes to sequentially numbered PNM files
		with one file for each frame (e.g. frame00000.pnm, 
		frame00001.pnm, etc.)

	-n
		Normalizes colors. Usually improves the saturation and color
		balance for old films with faded colors and scans with 
		inadequate gamma correction. However, this can introduce
		additional flicker since normalization will be affected
		by dust, scratches and inconsistencies in scanner source 
		light over the length of the scan

	-resolution [dpi]
		Sets input file resolution to given DPI. Default is 1200

	-l [input_file]
		Lists the chunks (.avi) or atoms (.mov) in the given file. Does
		not do any further processing and stops after list. 
		This feature is provided for diagnosing problems or
		examining the content of existing .avi or .mov files

	-rate [fps]
		When output file is given, sets the display rate in frames
		per second. Default is 18FPS (silent film standard)

	-b 
		Output files are converted to grayscale. 

	-h
		Prints this syntax message

	-clipleft [pixels]
	-clipright [pixels]
	-cliptop [pixels]
	-clipbottom [pixels]
		
		Clips the given number of pixels from the edge of the output frame.
		These values can be negative. Used for cropping.

	-framenumber [number]
		When output is going to separate files (not .avi or .mov), this
		sets the starting number used for the file names. Default is 00000

	-nosprocket
		Turns off sprocket hole detection and assumes the top of the first
		frame is the top of the input image(s) and that subsequent frames
		follow at exactly the appropriate frame height for the given resolution.
		Useful for scans of commercial prints where sprocket holes are
		not detectable in the scan. 

	-frameoffset [number]
		When not using sprocket hole detection, the top of the first frame
		is assumed to start at the very top of the scan file. This 
		parameter can be used when the first frame should start at
		a lower location. Value is in pixels and can only be positive.
		Note that this parameter turns off sprocket hole detection.

	-frameheight [number]
		When not using sprocket hole detection, variation in scanner
		accuracy will cause frame drift. This parameter can be tweaked
		to adjust the spacing between frames. Values are in pixels.
		Default is 200 at 1200 DPI. This parameter turns off sprocket
		hole detection: -cliptop and -clipbottom should be used for 
		adjusting framing and/or modifying the size of the output frame.
		
	-framefromright
	-framefromleft
		Normally a dark left edge is assumed for finding the horizontal
		location of a frame. This should also work for commercial
		prints that have clear sprocket area since the detected width
		will be less than 8mm and the left edge is adjusted accordingly.
		However, some commercial prints are framed so the picture area
		abuts the clear sprocket area. If the content of the picture 
		is light, there will be no left edge to detect and horizontal
		framing will be incorrect. In those cases, there may be a dark
		right border which can be used. -framefromright causes usage
		of the right border for framing. The default is framing from
		the left and -framefromleft can be used to restore left
		framing if -framefromright was given for a previous file in a 
		list of input files.
		
	-sprocket
		Turns sprocket hole detection back on if -nosprocket was given
		before a previous input file.

Examples:

Home Movies: Home movies normally have emulsion covering the entire width of the film and therefore have white sprocket holes against a black background. In this scenario, when given the correct resolution, mmusuper8 should be able to detect sprocket holes and create frame files without any additional parameters. Multiple input strip files can be included with each invocation of mmsuper8. The following invocation assumes the strip files are scanned at 1,200 DPI and will create a sequence of files named frame00001.pnm, frame00002.pnm, etc.

	mmsuper8 strip001.pnm strip002.pnm strip003.pnm

Commercial Print: Commercial super 8 prints often have emulsion only in the frame area of the film. Therefore, the sprocket area is clear and mmsuper8 is incapable of detecting sprocket holes. Complicating matters, the printed image may not be correctly centered in the frame area, making it impossible to detect the left edge of the film. In these cases, specific framing information FOR EACH STRIP FILE must be given.

	mmsuper8 -nosprocket -frameoffset 280 -frameheight 397 -framefromright 
		-clipright 20 -clipleft 20 -resolution 2400 2400_strip.pnm -o temp.mov

The example above works with this brief sample strip. Note that this strip is in PNG format to save server space since PNM files are quite large. It needs to be converted from PNG to PNM with gimp or some other image processing utility. An brief example of this strip converted to a movie (in RealVideo format) is given here.

-nosprocket: turns off automatic sprocket hole detection
-frameoffset 280: indicates that the top of the first frame in the strip starts at y pixel 280. Since it is very difficult to accurately scan the film with the top of the frame matching the top of the scan, this will probably change with each strip scan file.
-frameheight 397: At 2,400 DPI, frames should be 400 pixels high. However, scans are rarely perfect and frameheight must be tweaked so the framing does not shift as mmsuper8 progresses down the strip. Note that this does not adjust the actual height of the extracted frame; cliptop and/or clipbottom must be used for that, although the default is usually fine.
-framefromright: This particular print has a poorly centered frame and the frame bleeds over left edge, which is normally used to find the edge of the film. As such, when the frame content is lightly colored, it is impossible to detect where the edge starts. However, the right edge is consistently dark, so this flag indicates to use the right edge
-clipright 20 -clipleft 20: The mmsuper8 default frame width will normally capture the entire frame area width. However, in this example, because the left and right edges of the film are clear and because of the the poor frame centering of the print, the left and right edges are cropped to remove extraneous material.
-resolution 2400: mmsuper8 defaults to 1200 DPI and will not attempt to infer DPI from the scan file since PNM file headers do not include any DPI information. -resolution MUST be specified if the scan is anything other than 1200 DPI.
-o temp.mov: Specifies the output should be written to a Quicktime (.mov) file.

Converting Frames to Quicktime: Although it is useful for initially converting scan strips to individual frame files for editing or archiving, the final viewable product will need to be some video format. mmsuper8 can be used to combine frame files into uncompressed Quicktime files:

	mmsuper8 frame*.pnm -resolution 2400 -o final_movie.mov

The example above uses a wildcard to specify multiple frame files. If the frame files are anything other than 1200 DPI, -resolution must be specified. There appears to be a size limit on .mov files (that mmsuper8 does not detect), which limits the size of uncompressed movies created with high-resolution files to only a few seconds. In those cases, it is probably best to create a multiple short .mov files, combine them with some video editing software and then save the composite file in a compressed format that uses space more efficiently.

5. Playback and Transfer

The mmsuper8 package include mmsuper8play, a simple X-Windows player for playing individual PNM frame files. Since this reads files on demand and does not attempt to sync with the refresh rate of the display, playback rate is inaccurate and unstable. However, it is useful for previewing and editing frame content prior to conversion to .mov or .avi. mmsuper8play uses the following keystrokes:

Left Arrow: decreases frame rate by 3 FPS. This can also move to a negative rate that plays backwards.
Right Arrow: increases frame rate by 3 FPS.
Space: Pauses/restarts playback
Home: Go to beginning frame
End: Go to last frame
PgUp: Advance one second (18 frames)
PgDown: Back up one second (18 frames)

On Linux, the .avi players play with mplayer. The .mov files play with lqtplay, a simple player that is part of the libquicktime package. The .mov files do not play on mplayer.

Although they conform to the published file formats, the video output files created by mmsuper8 do not play on the Windoze media player (.avi), Widoze Quicktime player (.mov) or the RealPlayer on any platform. Players on Windoze are closed source and the error messages issued by the players are too inspecific to diagnose the problem further.

RealProducer can import .mov files and output compressed .rm files that can be streamed on the internet. You can download a Linux tarball from the Real.com website.

I assume that digital video editing software can import .mov or .avi files, although I have never tried.

5. Utilities

MMSUPER8 contains two utility programs for modifying frame files. Both these utilities operate on separate frame files, not a single strip containing multiple frames. Neither utility is fully developed.

mmsuper8stabilize removes inter-frame jitter and flicker. It works by performing correlation between frames and shifting frames to achieve maximum correlation. As such, it is really only useful on scenes that are largely static. It also most effective only when used on frames that constitute a single scene; it should not be run on frames that constitute multiple scenes. It does not crop the edges which will will look strange with extraneous material.

	mmsuper8stabilize max_shift pnmfile1 pnmfile2 ...

		max_shift is in number of pixels.

mmsuper8jitter adds jitter, flicker and grain to a sequence of frame files to create the illusion of an old movie. This should be used on files with individual frames, NOT scan files that contain strips of frames.

	mmsuper8jitter [options] [pnmfile1 pnmfile2 ...]
    
		-jitter (pixels)
			Simulates worn sprocket holes which cause steady 
			images to jitter slightly. Specifies the maximum 
			value for random frame position shift between frames. 
			Default is 1 pixel. Larger values look more like
			an earthquake than jitter. Value of 0 turns off
			this effect.
		
		-flicker (level)
			Simulates random variability in image intensity
			caused by inconsistent film sensitivity and
			inconsistent frame exposure times. Specifies a 
			maximum value for random intensity value added to 
			each frame, assuming 8-bit images with an intensity 
			range of 0 - 255. Default is 16. Value of 0
			turns off this effect.
		
		-grainlevel (grain level)	
		-graincount (grain count)
			
			Simulates film grain by adding random points of
			intensity to the image. Specifies the maximum intensity
			of each point (assuming 8-bit depth). Default level 
			is 12 and default count is width * height * 2.